1. Field of the Invention
The present invention relates to electronic systems and, more particularly, to debugging of electronic systems.
2. Description of the Related Art
Electronic systems are designed by designers to operate in specific ways. Electronic systems are systems that contain digital and/or analog electronic components connected together to perform specific operations or functions. Besides the electronic components, electronic systems may also include software. Once designed, the electronic systems may need to be debugged. Debugging electronic systems is a process which involves detection, diagnosis, and correction of functional failures. In the detection step, the designer of the electronic system observes a functional failure. When the designer is able to gather enough information about the incorrect behavior of the electronic system, the designer of the electronic system can draw the necessary conclusions to diagnose the functional failure. For correction of the functional failure, a fix is applied and subsequently tested. When the design is provided in a Hardware Description Language (HDL), such a fix may be a textual change to the HDL description of the electronic system.
In general, debugging has conventionally been performed by various different approaches. In particular, debugging has been performed by computer software debugging, hardware description language functional verification, hardware logic level analysis, or hardware behavioral source level emulation. These different approaches are discussed below.
Computer software debugging is conventionally performed using a computer software debugger. A computer software debugger is a software tool that allows a software developer to control the execution of a running computer software program by setting break-points, sequentially single-stepping through the execution of the computer software program, and looking at the program's state by examining and displaying variables and expressions. One example of such a software debugging tool is the GNU Debugger (GDB), which can be obtained from Red Hat, Inc. in Sunnyvale, Calif.
Software debuggers usually offer interactive debugging of software programs which are sequentially executed on computers. However, some software debuggers also support limited concurrency such as threaded program execution. Some software debuggers support debugging programs written at different levels of abstraction from high-level computer languages such as C++ down to assembler code and/or machine code. To assist with debugging of programs written in high-level computer languages, the software debugging system can add extra debug information (e.g., symbolic names and references to source code) to the compiled code during compilation of the computer software program. In combination with in-circuit emulators, software debuggers may provide a limited capability to analyze the underlying Central Processing Unit (CPU) of the computer executing the computer software program. A major disadvantage of software debuggers is, however, that they cannot be used for efficiently debugging general hardware of electronic systems.
Hardware description language functional verification is used to verify that the parts of an electronic system which are described using HDL match their functional specification. Such functional verification can be achieved through functional simulation or formal verification.
Functional simulation is performed by a functional simulator. A functional simulator is a software program that runs on a host computer and simulates the operation of an electronic system using its HDL description. Examples of functional simulators include VCS and VSS from Synopsys, Inc. in Mountain View, Calif., and ModelSim from Mentor Graphics Corp. in Wilsonville, Oreg. To increase simulation performance some functional simulators additionally make use of special purpose hardware which acts as a co-processor and accelerates the simulation. An example of a hardware-accelerated functional simulator is the Hammer system from Tharas Systems, Inc. in Santa Clara, Calif. Unfortunately, one major disadvantage of functional simulation is the need for simulation models. In order to be able to simulate, there must exist a simulation model with the proper functional behavior for each component of the HDL design for the electronic system. For some components such simulation models may not be readily available and must be generated. Additionally, the HDL design must be stimulated by a testbench. Since the ideal testbench must correctly and exhaustively match the behavior of the target environment, creation of a testbench can be very difficult and time consuming. On the other hand, a testbench that is too simple will not provide the necessary coverage to find all the design errors. Although functional simulation is useful, using functional simulation to debug design errors is too burdensome. Not only are the testbenches difficult to create, but also the more complex the HDL design is, the lower the performance of functional simulation. For state-of-the-art HDL designs simulation is now a million times slower than the fabricated hardware. Hardware-acceleration can typically speedup functional simulation by a factor on the order of one-hundred. Accordingly, its low performance makes it impractical to use functional simulation either to debug real-time applications or to concurrently debug hardware and software of complex electronic systems.
Formal verification is performed by a formal verification tool. Formal verification can help with the problem of incomplete coverage in functional simulation due to testbench limitations. One approach checks the HDL description for properties. Such properties may be explicitly provided by the designer of the electronic system or implicitly extracted from the HDL description by the formal verification tool. An example of such a formal verification tool is Solidify from Averant, Inc. in Sunnyvale, Calif. One disadvantage of formal verification is that it is impractical to use to re-produce functional failures observed in a running electronic system.
Both techniques, functional simulation and formal verification, have the major disadvantage that they do not operate on fabricated hardware. Instead, both techniques operate on a model of the electronic system and a model of the environment in which the electronic system runs, i.e., a testbench. Thus, their use is limited to debugging design errors. As such, neither technique is applicable for debugging manufacturing faults, environment errors, timing errors and/or tool errors. Also, inadequacies in the testbench have the potential to hide or introduce design errors in the HDL design during functional simulation which can later, when the HDL design is fabricated, show up as functional failures of the running electronic system.
Hardware logic level analysis is a technique that works at the logic level of a fabricated electronic system. The logic level of abstraction is also referred to as gate-level. Since electronic systems have been designed at the logic level for many years (for example using schematic entry of logic gates and flip-flops), there exists a wide variety of different techniques for debugging at logic level, including: digital logic is analyzers, in-circuit emulators, Design-For-Test (DFT) techniques, and hardware emulation, each of these different techniques are discussed below.
Digital logic analyzers operate to probe a limited number of digital signals and record their logic values. Probing is accomplished by physically connecting probes of the digital logic analyzer to exposed pins and/or circuitry on the fabricated design. Recording is controlled by trigger conditions, which are conditional expressions built upon the values of the recorded signals provided by the probes. The values for the recorded signals are stored in dedicated memory inside the digital logic analyzer so as to be available for subsequent display. Digital logic analyzers can be external devices or blocks embedded inside the digital circuits of an electronic system. An example of an external digital logic analyzer is the Agilent 16715A from Agilent Technologies, Inc. in Palo Alto, Calif. Examples of embedded logic analyzers are SignalTap from Altera Corporation in San Jose, Calif., or ChipScope from xilinx, Inc. in San Jose, Calif. Another example of an embedded logic analyzer was presented at the 1999 IEEE International Test Conference by Bulent Dervisoglu in “Design for Testability: It is time to deliver it for Time-to-Market”. Since embedded logic analyzers are added to the circuitry of the design, they can probe internal signals. Thus, embedded digital logic analyzers overcome the limited access to internal signals problem of external logic analyzers because access to the internal signals is not restricted by the pins of the fabricated circuits.
An in-circuit emulator is a specialized piece of hardware that connects to a CPU for debugging the CPU and the software that runs on the CPU. An example of an in-circuit emulator is visionICE from Windriver in Alameda, Calif. However, since in-circuit emulators only work for the specific target CPU for which they were built, in-circuit emulators are inappropriate for debugging general digital circuits.
DFT techniques, such as boundary scan and built-in self test, provide access to the internal registers of a running fabricated digital circuit. An example of such technique is described in the IEEE 1149.1 JTAG standard available from the Institute of Electrical and Electronic Engineers in Piscataway, N.J. DFT techniques are also described in “Digital Logic Testing and Simulation” by Alexander Miczo, published by Wiley, John and Sons Inc., 1985. DFT techniques were originally developed for and applied to testing of manufacturing faults and have the major disadvantage that they do not relate back to the HDL description.
Hardware emulation systems map a synthesized HDL design onto special emulation hardware. Such emulation hardware comprises many re-programmable FPGA devices and/or special purpose processors. The emulation hardware then executes a model of the HDL design. Thus hardware emulation has the same disadvantage as functional simulation, namely, that it works on a model of the electronic system and not on the fabricated hardware. As a result, hardware emulation systems are limited to design error debugging, and cannot be used for diagnosing manufacturing faults, tool errors, timing errors, etc. An example of such a hardware emulation system is System Realizer from Quicktum Systems, in San Jose, Calif. Specially built prototyping systems comprising FPGAs/PLDs can also be seen as hardware emulation systems. Since hardware emulation is usually much faster than functional simulation, hardware emulation systems may enable use of the software that is supposed to run on the HDL design to be used as a testbench. Even so, hardware emulation typically runs at speeds below one MegaHertz (MHz) while the HDL design is supposed to run at many hundred MegaHertz. In some cases the emulator speed may allow the user to connect the HDL design to the target environment which makes the design of testbenches unnecessary. Even so, with the high speeds of state-of-the-art HDL designs, hardware emulation is not capable of debugging the majority of real-time applications. Another disadvantage is that the special synthesis, mapping, and multi-chip partitioning steps required to bring an HDL design into a hardware emulation system are very complicated and time consuming.
A major drawback of all logic level debugging techniques is that they work at the logic level of abstraction. Since the HDL-based design methodology of electronic systems is much more efficient for todays complex designs, HDL designs have largely replaced logic level designs. Application of logic level debugging techniques to HDL design methodology is highly inefficient. Since logic level debugging does not-relate back to the HDL description, it normally would not provide the designer of the electronic system with sufficient information to correctly diagnose a functional failure.
Hardware behavioral source level emulation provides hardware emulation of source level designs. One technique for debugging HDL designs described at the behavioral level HDL using hardware emulation is described in “Interaktives Debugging algorithmischer Hardware-Verhaltensbeschreibungen mit Emulation” by Gemot H. Koch, Shaker Verlag, Germany, 1998. Some of which is also described in Koch et al., “Breakpoints and Breakpoint Detection in Source Level Emulation,” ACM Transactions on Design Automation of Electronic Systems, Vol. 3, No. 2, 1998. The therein described technique is referred to as Source Level Emulation (SLE) and offers an approach for emulating HDL designs, however only if such designs are described in behavioral VHDL. During behavioral synthesis a behavioral HDL design is enhanced for debugging by generating and adding additional circuitry for break-point detection. The behavioral synthesis tool writes out synthesized VHDL which contains a register transfer level description of the enhanced HDL design. The register transfer level description is then synthesized, mapped, and multi-chip partitioned into the emulation hardware. During hardware emulation with a hardware model of the HDL design, the user is able to examine particular variables in the behavioral HDL description.
Control is provided via break-points which are detected using the additional circuitry inside the running hardware model. Break-points in SLE have a very specific meaning. In particular, such break-points are closely tied to behavioral operations in the data-flow of the behavioral HDL description, and are associated with particular states of a controller which is generated by the behavioral synthesis. Additionally, break-points can be made conditioned upon particular values of data-path registers. When a break-point is detected, the execution of the hardware model is stopped. This is done by halting some or all of the system clocks and prevents the registers from changing their current values. Once halted, internal registers can be read. These registers form a scan-chain such that their values can be read by an emulation debugging tool.
Examination of variables in the behavioral HDL description is done in two ways. For variables which are mapped by the behavioral synthesis into registers in the hardware model, their values can be read and related back to HDL identifiers. This is done using map files which keep track of the transformations in behavioral synthesis, register transfer level synthesis, mapping, and multi-chip partitioning. For variables which have not been mapped to registers in the hardware model, their values are computed using a functional model of the behavioral HDL design. This functional model is created during behavioral synthesis and requires the existence of functional models of its components. The values, either read or computed, are then displayed in the behavioral HDL description. Optionally, by overwriting some or all of the registers of the hardware model while the hardware model is halted, the behavior of the HDL design can be modified once the execution of the hardware model is resumed.
Although source level emulation provides a debugging method which works at the level of the HDL description (in this case behavioral VHDL), it has various drawbacks which limits its use in practice. Several of the drawbacks are as follows. First, enhancements for source level emulation must be done inside a behavioral synthesis tool, since it needs special information about the behavioral HDL design which is only available during the behavioral synthesis process. Second, source level emulation does not allow the designer to perform customization. For example, a designer is not able to select trade-offs between hardware overhead and debugging support. Third, source level emulation cannot handle HDL descriptions on levels of abstraction other than the one provided by behavioral VHDL. Explicitly, source level emulation is not applicable for the most commonly used levels of abstraction of RTL HDL and gate-level HDL. Fourth, source level emulation supports neither hierarchy nor re-use of pre-designed blocks. Fifth, there are various limitations and difficulties in relating registers back to behavioral HDL source code. Sixth, in order to examine the state of the hardware model, it is required that some or all of the system clocks be halted and the hardware stopped, which makes source level emulation inapplicable for debugging the majority of today's electronic systems which are not to be stopped.
Thus, there is a need for efficient and effective approaches for debugging HDL-based electronic system designs.